Highly-Robust Feature Detection Method in Shape-Coded Structured Light Based on End-to-End Deep Neural Network

Shape-coded structured light is one of the most important technologies in the field of structured light. However, due to its poor robustness in feature extraction, its application in practical scenes is greatly limited. To solve this problem, this paper developed a highly-robust feature extraction method based on the deep learning method. First, a shape-coded structured light pattern was designed. The grid points formed by a series of horizontal and vertical lines were designated as the coding feature points. Then, the famous U-net network was utilized to detect the location of the grid feature points. At last, to varify the effectiveness of the proposed method, a large number of experiments were carried out. Experimental results showed that our method outperforms traditional feature detection method, especially in the case of noise interference. For structured light pattern with higher coding density, our method performs much better than the traditional method in both detection robustness and detection accuracy.